I am working on a survival analysis using the RMA normalized/transformed data of some cancer patients.

Here is a quick overview on RMA: http://www.molmine.com/magma/loading/rma.htm

As mentioned in the link above, RMA is already log-2 transformed and quantile normalized(skewed), so when I tried to calculate z-scores using these RMA values, it was not working (I am not getting a mean of zero when I check the summary of the global z-scores, although I have re-calculated them multiple times). Hence, I did my survival analysis (Cox regression) with the RMA values itself.

But the problem I am now running into is that without calculating z-scores, I am not sure how to establish what constitutes of high/low expression using just the RMA data, which I need to do in order to create Kaplan-Meier curves and do further analysis. Do I establish this cutoff/range using the mean/median along with the hazard ratio, or is there some other more statistically accurate way?

Any help would be greatly appreciated, thank you!

Dr. Blighe - thank you for the response. I have clarified what I mean by "not working", sorry for the ambiguity. If I am dividing the RMA into expression ranges of tertiles, quartiles, etc., that is something I can do after the Cox regression and need only when I am plotting the survival graph, correct? Or would I have to redo everything I have done so far with the Cox regressions?

You can user either approach. There are no strict rules regarding this.

Ok, thank you for all the help!